v2.1.1

A correctness pass over what the run reports. No input or output
changes, and the build verdict is untouched: the scanner still owns it.

Fixed

• Annotations followed a hard-coded critical/high list, so a run
  configured with fail-on: wcag failed the build with an empty
  annotations panel: Level A findings that gate are frequently medium.
  The annotated set is now whatever this run's gate can fail, read from
  the gate itself, and framework-managed findings are never annotated
  because they cannot fail the build they would sit beside.
• The comment's counts could exceed the verdict beside them. They
  came from summary.severity_counts, which counts every headline row,
  while the gate is evaluated on violations, which excludes the
  framework-managed findings the backend deliberately skips. The comment
  now prefers the gate's own numbers and falls back to the summary for
  older backends.
• Copy-as-PR was offered to runs that can never use it. Fixes are
  generated for scans attributed to an account; an anonymous scan has no
  account that could open the tab. The link now appears only on an
  attributed scan, and an anonymous one is told what to add instead of
  being sent to a tab that will not serve it.
• The free-tier permissions note overstated the risk. Without
  id-token: write, quota is bucketed on the repository name the
  workflow reports, which nothing verifies; it is not shared across
  every repository behind a runner IP. Adding the permission makes the
  bucket provable rather than merely private.

Documentation

• plan-tier and ci-scans-remaining, added in 2.1.0, are in the
  outputs table.
• The accessibility-pro-token description says what a token actually
  buys: attribution, which is what lets you open the report signed in
  and generate Copy-as-PR fixes.